Deep Learning-Based Protein Half-Life Prediction for Identifying Rate-Limiting Enzymes in Metabolic Pathways to...

Area of Science:

• Synthetic biology
• Metabolic engineering
• Computational biology

Background:

• Enzyme stability is critical for efficient synthetic metabolic pathways.
• Short-lived enzymes limit metabolic productivity due to low abundance.
• Predicting protein half-life in bacteria is challenging.

Purpose of the Study:

• To develop a machine learning model for predicting bacterial protein half-life.
• To identify rate-limiting enzymes in synthetic metabolic pathways.
• To enhance metabolic productivity through targeted enzyme stabilization.

Main Methods:

• Developed ProHL, a multimodal machine learning model.
• Integrated ProteinBERT and physicochemical encodings for feature extraction.
• Validated ProHL on an independent dataset of E. coli proteins.

Main Results:

• ProHL achieved 0.818 accuracy and 0.624 Matthew's correlation coefficient.
• Identified CrtB as a short-lived enzyme in lycopene biosynthesis.
• Overexpression of CrtB increased lycopene production by 25% in E. coli.

Conclusions:

• ProHL effectively predicts bacterial enzyme half-life.
• In silico prediction of enzyme stability can alleviate metabolic bottlenecks.
• This approach offers a viable strategy for enhancing metabolic productivity.